Predicting Anticancer Drug Response With Deep Learning Constrained by Signaling Pathways.

Area of Science:

• Computational biology
• Genomics
• Pharmacology

Background:

• Precision medicine leverages multiomics data for tailored cancer treatments.
• Predicting drug response from complex, heterogeneous multiomics data and understanding the underlying mechanisms remain challenging.
• Existing computational models struggle to fully elucidate the molecular mechanisms of drug response.

Purpose of the Study:

• To develop an interpretable deep learning model for anti-cancer drug response prediction using signaling pathways.
• To integrate multiomics data (gene expression, copy number variation) within a pathway-constrained framework.
• To identify the importance of specific signaling pathways in predicting drug response.

Main Methods:

• A novel deep learning model, consDeepSignaling, was developed, incorporating 46 signaling pathways to constrain predictions.
• The model was trained and evaluated using multiomics data from the Cancer Cell Line Encyclopedia (CCLE) and drug response data from the Genomics of Drug Sensitivity in Cancer (GDSC) databases.
• Model interpretation techniques were employed to analyze the importance of signaling pathways.

Main Results:

• The consDeepSignaling model demonstrated superior performance compared to existing deep neural network models in predicting anti-cancer drug response.
• The model successfully integrated diverse multiomics data types.
• Pathway importance analysis revealed distinct patterns associated with drug response.

Conclusions:

• Pathway-constrained deep learning offers a powerful approach for enhancing the accuracy and interpretability of multiomics-based drug response prediction.
• This method advances our understanding of the molecular mechanisms driving anti-cancer drug efficacy.
• The findings support the potential of consDeepSignaling for advancing personalized cancer therapy.